|
Page 2 of 5 69 polls in all, including the latest Roy Morgan poll (60.5/39.5). Discussion on the latest poll is happening here (Oz Election 20007 forums) .
| | | L/NP | ALP | Others | | L/NP | ALP | | All 69 polls | Dec - Sep | 37.61 | 48.46
| 13.93 | TPP | 43.2 | 56.8 | | 1st set of 23 polls | Dec - Apr | 36.61 | 48.31 | 15.08 | TPP | 42.6 | 57.4 | | 2nd set of 23 polls | Apr - Jul | 37.75 | 48.72 | 13.53 | TPP | 43.2 | 56.8 | | 3rd set of 23 polls | Jul - Sep | 38.44 | 48.35 | 13.21 | TPP | 43.7 | 56.3 |
 Monthly Averaged (TPP @ 60% flow from Others to ALP): | | | L/NP | | ALP | | Others | | L/NP | ALP | June
| | 39 | | 47.3 | | 13.7 | TPP | 44.5 | 55.5 | | July | | 38.9 | | 47.3 | | 13.8 | TPP | 44.4 | 55.6 | | Aug | | 38.5 | | 47.9 | | 13.6 | TPP | 43.9 | 56.1 | | Sep | | 37.8 | | 49.6 | | 12.6 | TPP | 42.8 | 57.2 |
Some Polling Explanations:
Think for a moment you have a bag of 100 balls, 50 black and 50 white. This you know for a fact. You randomly select 20 balls from the bag. For your sample to accurately reflect the 50/50 split between the white balls and the black balls, you should have pulled out 10 white balls and 10 black balls. But you wouldn't at all be surprised if you pulled out only 5 white balls and 15 black balls. You would put it down to a chance variation in the sample, you wouldn't immediately assume that the bag did not originally have 50 white and 50 black, and that the make up of balls has changed to 25 white and 75 black, to reflect the sample. Equally, if after replacing the first sample, your next sample pulled out 15 white balls and only 5 black, you wouldn't conclude that there has been a swing back to the white balls.
If we repeat the sampling exercise many times - taking out 20 balls and recording the score, then replacing the balls and then taking out another 20 balls - keep doing it enough and you would get a very good representation of what's in the bag and your results would very closely approximate 50 white balls and 50 black balls. Of course, in this example, you could just take out all 100 and count them up once, but that's what is called a census.
When you have a population of 13 million voters, you can't do a census every time you want to research voting intentions, that is done once every three years at the election. All you can do is take a random sample of the voters and ask them how they are intending to vote. Just as in the example with the balls; where you wouldn't believe that one sample from the bag accurately reflected what was in the bag; nor should you with one poll of voting intentions. You become more confident with two polls, three polls, four polls and so on.
What happens with political polls is that the media look at the difference from one poll to another, even if they are taken over the same weekend, and attribute all sorts of reasons to explain the difference. Where the truth is, nothing has happened. It's just like the balls, the make up of 50 white and 50 black hadn't changed, but that's exactly what the media argue. They truly believe that something in the week explains the difference in the results.
However, there is one difference between pulling balls out of a bag and estimating voting intentions, and that is that voting intentions may change over time. That is why it's best to look at polling data over time to see if there is indeed any change in voting intention, and if there is, the trend should show up pretty clearly.
The only other thing that is often suggested, is that polling is not actually measuring voting intentions, but rather something else. What that something else is, no one can actually define. But this argument really doesn't stack up, because the question respondents are asked is straightforward and unambiguous.
So the next time a poll is released, think of the balls in the bag.  - Aristotle (data, comment), JJ (layout)
Views: 451
|
Comments (24)
